When the model is the policy, the diff is the platform.
The objectives, output schemas, and optimization targets of public models are value-laden decisions — not technical ones. What does a "good permitting outcome" mean? Speed? Equity? Environmental compliance? These questions belong to the public. Participatory input processes — voting mechanisms, public comment, citizen assemblies — set the foundation before representatives or technologists touch the implementation.
Future elected officials will campaign on specific, auditable proposals: what they would change in the model, what objectives they'd reweight, what outputs they'd require. Unlike today's vague policy promises, these are diffs — verifiable, comparable, and trackable. Did the representative's changes produce the outcomes they promised? The record is in the model's history.
Explainability requirements — what changed and what the downstream effects were — are civic infrastructure, not technical nice-to-haves.